I’m a senior undergraduate at National Taiwan University, majoring in Electrical Engineering with a double major in Physics. During Fall 2026, I’m an exchange student in Electrical and Computer Engineering at UC San Diego.
My research interests lie in generative modeling and machine learning for spatiotemporal and scientific data. I currently work with Prof. Rose Yu (UC San Diego) on context-aware spatiotemporal forecasting, Prof. Han Liu (Northwestern) on generative models for single-cell perturbations, and Prof. Hung-Yu Wei (NTU) on generative models for wireless time series.
I also enjoy building end-to-end ML systems, from LLM-powered analytics platforms to real-time brain–computer interfaces.
Advisor: Prof. Rose Yu
Designing a spatiotemporal forecasting benchmark that evaluates where, when, and how real-world events affect economic regions connected by a graph. Curated 22 economic and financial time-series sources and built a USDA grain-logistics graph with 19 markets and 36 state-to-port edges.
Advisor: Prof. Han Liu · Remote
Developing PyTorch generative models that predict single-cell gene-expression responses to genetic perturbations, and studying the optimal mix of synthetic and real training data. Built a sparse-matrix QC and donor-level aggregation pipeline across 2 studies (25,690 selected cells, 20,889 shared genes).
Advisor: Prof. Hung-Yu Wei
Developing deep generative models that synthesize wireless time-series data for digital twins. Built 199,567 overlapping sequence windows from 791 wireless trace files and adapted TimeVAE for wireless data synthesis.
Advisor: Prof. Yu-Chiang Wang
Surveyed video diffusion-based world models, analyzing how architectures support autoregressive rollout and spatial memory.
An AI assistant for semiconductor supply-chain insights. LLM pipelines turn 300+ news articles and 100 earnings-call transcripts into summaries and risk profiles for 20 companies; a Vertex AI agent with 4 retrieval tools over BigQuery and Cloud Storage answers company-grounded questions.
A brain–computer interface game. A CNN-Transformer (CTNet) classifies focus vs. relaxation from 500 Hz EEG, reaching 71.3% mean leave-one-subject-out accuracy across 35 subjects (baseline 55.1%), and streams predictions over TCP to steer a physics-based balance game.
Classifies Alzheimer’s disease, frontotemporal dementia, and healthy controls from 19-channel EEG of 88 subjects. A two-stage SVM on graph-signal features reached 74.1% accuracy and 0.767 macro ROC-AUC on a 27-subject test set, benchmarked against random forest, logistic regression, and a CNN-Transformer.
A Flutter app that streams Arduino ECG/EMG measurements over Bluetooth Low Energy and displays live waveforms with heart-rate estimation.
A web app that converts sketches and Raspberry Pi camera photos into G-code through a custom JPG → SVG → G-code pipeline, then drives a pen plotter via Universal G-code Sender.
Exchange Student, Electrical and Computer Engineering
B.S. in Electrical Engineering, Double Major in Physics · GPA 4.12/4.30
Coursework: Machine Learning, Deep Learning, Computer Vision, Probability and Statistics, Linear Algebra, Signals and Systems, Algorithms, Data Structures, Web Programming
Short-Term Student, Taichung First Senior High School Science Program
Took the program’s 12th-grade coursework at NYCU.
Science Program
